Searcharxiv⌕ Search

arXiv subjects

Zhi Chen

Publications and source records attributed to Zhi Chen.

At least 37 records · Page 2Linked to original sources
arXiv2026

TCPO: Turn-Level Credit Policy Optimization

Verifier-guided reinforcement learning has become a powerful paradigm for improving LLM reasoning. In multi-turn settings, models receive a verifier score after each turn and iteratively refine their outputs. Although such scores provide dense feedback, they do not directly provide dense credit: a score measures the quality of the current output, while credit should measure how the current turn changes the refinement trajectory. We propose TCPO, a turn-level credit assignment method for verifier-guided multi-turn RL. TCPO casts credit assignment as score-to-credit conversion and constructs turn-level advantages through reference-based comparisons: retrospective credit captures immediate progress and regression relative to the best prior state; hindsight delayed credit identifies non-improving turns with later payoff; and selective fixed-history counterfactual estimation refines high-surprisal turns under the same history. Experiments on math reasoning, code generation, and AppWorld agent tasks show that TCPO improves or matches the strongest baselines across model scales, task domains, and verifier types. TCPO achieves the best or tied-best best-turn Pass@8 on Qwen3-4B and DeepSeek-R1-Distill-Llama-8B, reduces turns to success, and improves multi-turn agent performance. These results highlight score-to-credit conversion as a central ingredient for verifier-guided multi-turn policy optimization.

cs.AI↗
arXiv2026

Beyond the Best Teacher: Expanding and Compressing the Reasoning Solution Manifold

A single reinforcement-learning run can produce a strong reasoner yet an incomplete teacher: it often amplifies only a subset of the valid solution modes. We argue that reinforcement learning (RL)-trained policies should therefore be viewed as local probes of a multi-basin reasoning solution manifold, rather than as globally reliable supervisors. Based on this view, we propose an expand-then-compress framework that couples teacher construction with multi-teacher policy distillation. In the expansion stage, Residual Group Relative Policy Optimization (RGRPO) trains a sequence of teachers from a common initialization and redirects each later round toward examples not yet covered by the accumulated teacher union. In the compression stage, reliability-gated Teacher-Union On-policy Distillation (TU-OPD) lets the student learn from its own response prefixes. For each example, only reliable teachers contribute, and their sampled-token OPD losses are weighted by their per-example quality. We further introduce Consensus-Residual Decomposition, which preserves a winner teacher's excess token preferences over its reliable peers, preventing specialist behavior from being suppressed during teacher aggregation. Experiments on mathematical reasoning, code generation, and instruction following show that the resulting Qwen3-1.7B student consistently outperforms the strongest individual teacher across all three domains, yielding relative improvements of 2.0%, 8.3%, and 6.9%, respectively, while retaining single-model inference. These results establish a simple but powerful principle: stronger students can be obtained not by selecting a single better teacher, but by deliberately constructing and compressing a complementary teacher union.

cs.LG↗
arXiv2026

From Understanding to Action: Feedback-Grounded Policy Discovery for Generative Recommendation

Semantic-ID-based generative recommenders enable efficient next-item generation, but their item-level supervision mainly captures behavioral co-occurrence and local transitions. Large language models (LLMs) can complement these models by reasoning over heterogeneous interaction histories to understand the user's current demand. However, LLMs are not inherently trained with recommendation-specific outcome feedback, and linguistically plausible reasoning therefore does not necessarily lead to effective recommendation decisions. We term this mismatch the Understanding-Action Gap. Accordingly, we distinguish intent knowledge, which captures the user's current demand, from policy knowledge, which specifies the recommendation direction and rejection boundary under that demand. To bridge this gap, we propose a feedback-driven agent framework that first induces task-oriented intent and then discovers recommendation policies according to their incremental utility over an intent-only baseline. Candidate policies are evaluated and refined using outcome-derived feedback rather than linguistic plausibility. We further transfer the resulting intent and policy knowledge into two latent tokens of a lightweight Semantic-ID generator through dual-space relational distillation, enabling LLM-free online inference. Experiments on public benchmarks show consistent improvements over baselines, while large-scale online A/B tests achieve gains of 4.506% in Revenue and 4.621% in ADVV.

cs.IR↗
arXiv2026

MemTX: Transactional Belief Commit for Stateful Agent Memory

LLM agents increasingly coordinate through persistent shared memory: one agent's write becomes another agent's premise, and eventually a tool call with real side effects. Current agent memory systems treat every accepted write as immediately actionable truth, so a polluted tool result, a stale update, or a teammate's half-finished note can silently drive an irreversible action. We argue that a memory write is not a belief commit. We present MemTX, a transactional belief-commit protocol. Each record carries evidence, permissions, provenance, and validity. Writes are staged inside snapshot-isolated transactions and admitted by a validate-and-commit pipeline, irreversible tool calls are gated on in-flight belief state, and retracting a belief triggers typed cascading repair of its derived records and tool side effects. Two invariants, action-safety gating and cascade-repair completeness, are machine-checked by property-based testing and bounded exhaustive enumeration of 5.5 million protocol states, with zero violations. Across five backbones from three model families, MemTX leads all eight baselines with paired-McNemar significance on four backbones and statistically ties the best baseline on the fifth and strongest, while remaining the only method with zero downstream harm on every backbone. Backbone capability does not substitute for commit discipline.

cs.AI↗
arXiv2026

Quiver data of representations of Hecke algebras and Coxeter groups

Let Gamma be a Coxeter matrix and q be a generic parameter, let Hq (Gamma) be the Iwahori-Hecke algebra. We propose a concept H-representation of type [Gamma;q], which are representations of certain quiver satisfying several sets of relations. We show type [Gamma;q] H-representation and Hq (Gamma)-representation almost determine each other. By using this we construct all nonintersecting type Hq (Gamma)-representations . H-representation is compatible with the Kazhdan-Lusztig theory . Some relations between H-representation and W-graph are proposed, which produce an equation system for a W-graph to giving a W-representation of Hq (Gamma) . With the intention of constructing general type Hq (Gamma)-representations we introduce the concept H^2 -representation , which is another quiver representation hidden in a H-representation. A program to construct general type H-representations from a chosen [Gamma;q]-realizable H^2 -representation is developed ,by using it we prove that if Gamma satisfies m_(i,j) >= 5 for any i not equal to j, then any H^2 -representation is [ Gamma;q]-realizable.

math.RT↗
arXiv2026

MedDiT4SR: Tri-Stream Joint Adaptation of Pre-Trained Diffusion Transformers for Medical Image Super-Resolution

Medical image super-resolution (MedSR) requires recovering fine anatomical structures from degraded observations while avoiding unsupported details introduced by generative priors. Large-scale pre-trained multimodal diffusion transformers provide strong visual priors, but their adaptation to MedSR remains non-trivial. In conventional ControlNet-style adaptation, the low-resolution (LR) image is processed as an external condition and injected into the denoising stream through one-way connections. Consequently, LR anatomical evidence cannot be jointly updated with the evolving denoising and semantic representations. We propose MedDiT4SR, a tri-stream adaptation framework that integrates the LR, noisy latent, and text representations into the same multimodal diffusion-transformer blocks. To complement global token interaction, we introduce a Super-Resolution Adapter (SR Adapter) that aggregates scale-dependent local tokens and suppresses interpolation-induced redundancy. We further propose a Semantic Alignment Refiner (SA Refiner) that calibrates local LR responses using prompt-conditioned semantic information. Experiments under both in-domain and within-modality cross-dataset settings demonstrate the effectiveness of adapting large-scale pre-trained DiT models to medical image super-resolution across diverse imaging domains.

eess.IV↗
arXiv2026

Continual Video-MLLM Adaptation over Evolving Domains

Video multimodal large language models have shown strong capability in video understanding, yet their adaptation to sequentially evolving domains remains underexplored. In real-world deployments, video data often arrives continuously from heterogeneous domains, requiring the model to acquire new domain-specific knowledge without overwriting previously learned capabilities. Existing continual learning methods typically rely on shared adaptation spaces, which can induce severe cross-domain interference and catastrophic forgetting. We propose Distribution-Aware Expert Routing, a parameter-efficient framework for continual Video-MLLM adaptation over evolving domains. DAER maintains domain-isolated lightweight experts while keeping the pretrained Video-MLLM backbone frozen, thereby decoupling domain-specific adaptation from the general multimodal knowledge of the pretrained model. To enable fine-grained specialization, we introduce an intra-domain distribution-aware routing mechanism that matches each input to expert-level prototype reservoirs using MMD. To address the absence of task identities at inference time, we further propose an inter-domain routing mechanism that performs prototype matching in a discriminative subspace for robust domain identification. In addition, we introduce adaptive domain merging to improve parameter scalability and adopt a two-stage optimization strategy to stabilize expert specialization during continual learning. We evaluate DAER by curating a domain-incremental benchmark built from ten VidQA datasets covering diverse visual environments and reasoning demands. Experiments on two strong Video-MLLM backbones show that DAER consistently outperforms prior methods.

cs.CV↗
arXiv2026

Are Performance-Optimization Benchmarks Reliably Measuring Coding Agents?

Repository-level performance-optimization benchmarks such as GSO, SWE-Perf and SWE-fficiency evaluate coding agents by applying patches to real repositories and comparing runtime against unoptimized baselines and official reference patches. Their leaderboard scores are increasingly used as evidence of coding-agent progress, but those scores can conflate runtime instability, benchmark-specific scoring rules, and how many tasks are already solved by at least one public submission. We audit these issues across the three benchmarks. First, we replay the official reference patches for 740 code optimization tasks across four common types of Google Cloud machines. Most benchmark tasks can be replayed, but their reference patches satisfy the original benchmark validity rules in every cross-machine replay for only 39/102 GSO tasks, 11/140 SWE-Perf tasks, and 411/498 SWE-fficiency tasks; SWE-Perf is especially fragile because many reference patches produce close-to-zero runtime changes. Second, we show that public submission rankings depend strongly on the benchmark scoring rule. Among eight public submissions shared by GSO and SWE-fficiency, the official rankings disagree on 9 of 28 pairwise submission comparisons, and SWE-fficiency's leaderboard scoring rule assigns the worst ten tasks overly high score weights of 58.5%-82.8%. Third, looking across 10 public submissions for each task, we find that at least one submission matches or beats the reference patch on 85.3% (384/450) of replay-valid GSO and SWE-fficiency tasks, and beats the unoptimized base code on 99.8% (449/450). Our study complements leaderboard scores by identifying tasks with more reliable performance signals, quantifying per-task score contributions, and exposing the remaining performance gaps that are hidden by aggregate rankings.

cs.SE↗
arXiv2026

Bending Beam for THz Wireless Networks: Fundamental, Design Issue, and Prototype

Bending beams, characterized by their non diffracting and self-healing properties in the near field, offer a new approach to bypass blockage in terahertz (THz) wireless communication and sensing. However, the investigations of bend ing beams in the context of wireless communications still remain at an early stage. This article provides a state-of-the-art review of the fundamentals and key application scenarios of bending beams in THz wireless communications and sensing. We first present and compare the existing beamforming design and practical hardware implementation methods for bending beams. Next, we discuss potential applications of bending beams in wireless communica tions and sensing and identify their associated challenges, such as blocked channel modeling, bending beam training, codebook design, etc. Finally, a hardware demonstration of bending beam over THz frequency bands is presented, validating the advantages of bending beam over conventional beamfocusing.

eess.SP↗
arXiv2026

Stabilizing effect of a background magnetic field on the 2D damped wave-type MHD equations

The stabilizing effect of a background magnetic field on electrically conducting fluids has been rigorously established for the standard MHD equations. This paper extends this theory to the more physically accurate damped wave-type MHD equations, where the induction equation is hyperbolic-parabolic and the velocity field has only vertical damping with no dissipation. These two features make the stability analysis harder than in the standard MHD setting. To overcome these difficulties, we design an energy functional exploiting the anisotropic structure, and discover a remarkable cancellation between the two most dangerous nonlinear terms by exploiting the full algebraic structure of the coupled system. As a consequence, we prove that any small perturbation near the background magnetic field is globally stable and establish optimal decay rates consistent with the 2D heat equation. To the best of our knowledge, this is the first rigorous stability result for the damped wave-type MHD equations near a background magnetic field.

math.AP↗
arXiv2026

Benchmarking Vision-Language Models for Microscopic Plant Image Understanding

Microscopic imaging provides essential visual evidence for studying plant biology and pathology at the cellular and subcellular levels. However, existing benchmarks on vision-language models primarily focus on macroscopic plant imagery, while the microscopic domain remains underexplored. To address this gap, we present PlantMicro, a comprehensive benchmark for evaluating vision-language models (VLMs) in microscopic plant imagery. PlantMicro integrates more than 5,000 images collected across diverse hosts, biological domains, and imaging modalities. Building on this diversity, we design a set of complementary tasks that capture different facets of microscopic image understanding. To support these tasks, we construct over 9,000 VQA pairs that systematically evaluate the capabilities of VLMs. Experiments on PlantMicro show that current VLMs struggle with fine-grained recognition and biologically grounded reasoning. For example, GPT-5 achieves 34.93% accuracy on the pathogen classification task, which is only modestly above the random-guessing baseline. The results highlight a significant gap in current VLMs' ability to comprehend plant microscopic images. PlantMicro provides a standardized foundation for advancing VLMs toward reliable and comprehensive microscopy-level plant understanding.

cs.CV↗
arXiv2026

Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction

Visual token reduction has emerged as an effective strategy for accelerating Multimodal Large Language Models (MLLMs). Many existing methods prune tokens by ranking text-visual attention scores. However, we show that attention is often dominated by a model-induced prior: even without textual instruction, MLLMs tend to focus on certain task-agnostic regions. Consequently, the attention scores of instruction-conditioned tokens are suppressed, increasing the risk that these tokens are discarded during pruning. To address this issue, we propose Prior-Corrected Token Reduction (PriorTR), a training-free token reduction method that explicitly separates task-conditioned attention from the model-induced prior. PriorTR estimates the attention map of the prior, and contrasts it with the task-conditioned attention distribution to measure the additional usable information contributed by each visual token. Importantly, PriorTR computes both the model-induced prior and the task-conditioned posterior within a single forward pass by introducing a null token that serves as an instruction-agnostic probe in the attention block. This design avoids duplicated propagation. Extensive experiments across multiple multimodal benchmarks and MLLMs demonstrate that PriorTR consistently improves the trade-off between accuracy and efficiency over strong training-free baselines, particularly under aggressive token budgets.

cs.CV↗
arXiv2026

Spectral Evolution-Guided Token Pruning in Multimodal Large Language Models

Reducing visual token redundancy is critical for accelerating Multimodal Large Language Models (MLLMs) without degrading cross-modal reasoning performance. Existing token pruning methods typically rely on single-layer signals, such as attention scores or token similarities, which overlook the cross-layer transformation of visual representations and may exhibit positional bias in multimodal token sequences. To address this limitation, we propose a training-free token pruning framework based on Cross-Layer Spectral Evolution (CLSE). Instead of measuring token importance from single-layer feature magnitudes, CLSE quantifies how token representations evolve across Transformer layers in the frequency domain. This evolution reflects the transition from high-frequency structural details to low-frequency semantic abstractions. We observe that tokens with stronger spectral redistribution across layers are more likely to be semantically active and should therefore be preserved. By modeling cross-layer token dynamics, CLSE provides a stable importance criterion that mitigates positional bias. Extensive experiments on both image and video benchmarks demonstrate that CLSE achieves a superior trade-off between efficiency and accuracy under aggressive token reduction. Across multiple MLLMs, CLSE reduces FLOPs, KV cache memory, and latency while maintaining competitive or improved performance.

cs.CV↗
arXiv2026

Executing as You Generate: Hiding Execution Latency in LLM Code Interpreters

Current LLM systems are increasingly equipped with a code interpreter that executes generated code to obtain results. This works serially: the model first generates the complete code, then an interpreter executes it. This sequential workflow leaves the executor idle during generation and the generator idle during execution, resulting in unnecessary end-to-end latency. Our key observation is that an LLM, unlike a human developer, emits code tokens left to right and does not backtrack over what it has already written. This makes it possible to start executing a piece of code while later tokens are still being generated. We formalize this parallel execution paradigm, modeling it as a three-stage pipeline of generation, detection, and execution, and derive closed-form latency bounds that characterize its speedup potential and operating regimes. We then present EAGER, a concrete implementation featuring AST-based chunking, dynamic batching with gated execution, and early error interruption. We evaluate EAGER across four benchmarks, seven LLMs, and three execution environments. The overlap mechanism hides almost all execution behind generation, reducing the non-overlapped portion of execution time by up to 99.8% and cutting end-to-end latency by up to 37.3% on error-free runs.

cs.PL↗
arXiv2026

A Smart Classroom Behavior Analysis Framework with a New Highly Congested Classroom Dataset

Student behavior detection is important for intelligent classroom analysis but remains challenging in large-class scenarios due to dense instance co-occurrence, asymmetric occlusion, depth-wise scale variation, and fine-grained semantic degradation in distant targets. Existing classroom behavior datasets and general-purpose detectors are insufficient to characterize and address these challenges. This paper constructs the Highly Congested Classroom Behavior (HCCB) dataset, containing 50,229 student behavior instances across seven categories: reading, writing, heads up, sleeping, looking around, bowing head, and using phone. HCCB provides a challenging benchmark that integrates dense distributions, severe occlusion, scale variation, and fine-grained behavioral semantics. To address these issues, we propose ODER-HSFNet, a YOLO-based detection framework tailored to highly crowded classrooms. At its core, ODER-HSFNet introduces three task-specific innovations: the Occlusion-aware Deformable Edge Rectifier (ODER), which strengthens boundary evidence under occlusion; the Hypergraph-State Spatial Fusion (HSSF) module, which integrates local structure enhancement, state-space contextual modeling, and high-order relation aggregation; and the Occlusion-Calibrated Detection Head (OCDetect), which suppresses low-quality Pre-NMS candidates and reduces false positives from occlusion boundaries and neighboring instances. Experiments on two classroom behavior detection datasets show that ODER-HSFNet outperforms mainstream YOLO-series methods, achieving 60.60%/80.12% mAP50:95/mAP50 on HCCB and 57.36%/74.65% on SCB-D3-S. Ablation studies further verify the effectiveness of the proposed design for highly crowded classroom behavior detection.

cs.CV↗
arXiv2026

Deep Learning-Empowered Movable-Antenna Position Optimization with Partial CSI

Movable antennas (MAs) are a promising technology to improve wireless data rates by dynamically adjusting their positions to avoid deep fading. However, finding the optimal MA positions requires full channel state information (CSI) for all possible locations within the movement region, creating massive channel estimation overhead. This paper proposes a deep neural network (DNN)-based learning framework to predict the optimal positions of multiple transmit MAs in a multi-user multiple-input single-output (MISO) system, entirely bypassing explicit channel estimation.First, we analyze a single-user MISO case, revealing a complex, highly nonlinear mapping between the optimal MA positions and the channel power gains from a specific subset of locations in the transmit region to the user. Because this mapping cannot be mathematically characterized for practical channel models, we train a DNN via supervised learning to capture it. The pre-trained DNN can then determine optimized MA positions in real-time relying only on partial power measurements from the transmit region.Extending this to multi-user scenarios is challenging due to complex rate expressions and the lack of globally optimal position solutions to use as training labels. To overcome this, we develop an unsupervised training framework that directly maximizes the multi-user sum-rate. This framework utilizes an attention-based architecture to extract latent features from the partial channel measurements and effectively manage inter-user interference. Simulation results show that our proposed approach achieves near-optimal performance in single-user systems and surpasses conventional CSI-based alternating optimization algorithms in multi-user environments.

eess.SP↗
arXiv2026

TimeVista: Exploring and Exploiting Vision-Language Models as Judges for Time Series Forecasting

High-quality time series forecasting is pivotal for real-world decision-making. However, traditional point-wise metrics often fail to reveal complex temporal patterns and align poorly with human intuitive preferences. While the ''LLM-as-a-Judge'' paradigm has revolutionized text evaluation by providing flexible, human-aligned judgment, its application to time series remains largely unexplored. In this paper, we leverage Vision-Language Models (VLMs) as judges for time series forecasting, harnessing their ability to comprehend time series plots grounded in textual information. Specifically, we propose a novel framework integrating micro- and macro-level judgments informed by contextual information to evaluate time series forecasting. To this end, we introduce TimeVista, a comprehensive VLM-as-a-Judge benchmark comprising 5563 time series samples paired with detailed evaluation rubrics. Extensive meta-evaluations demonstrate that VLMs are highly reliable judges, achieving significantly higher consistency with human preferences than conventional metrics. Building upon our benchmark, we comprehensively assess recent Time Series Foundation Models (TSFMs) under the VLM-as-a-Judge paradigm. Our results demonstrate that VLMs serve as robust and interpretable judges, providing a comprehensive, human-aligned standard for evaluating time series models.

cs.AI↗
arXiv2026

Unlocking the Potential of Movable Antennas: General and Practical Antenna Position Optimization

Recently, movable antenna (MA) has attracted wide attention in wireless communications due to its potential in enhancing wireless communication performance via local movement within a confined region. However, antenna position optimization (APO) has emerged as a major challenge for MAs, due to the lack of a tractable, analytical, and accurate channel model in terms of antenna positions. Although existing works have developed various algorithms for APO, most of them are based on simplified theoretical channel models, which limit their generality. To address this challenge, in this article, we present more general and effective APO algorithms for different purposes, categorized as continuous APO and discrete APO, respectively. Continuous APO is mainly applied for flexible array signal processing to boost large-scale communication performance, while discrete APO is applied for small-scale multi-path channel reshaping. Specifically, the discrete APO discretizes the antenna movement region into multiple sampling points and employs discrete algorithms to determine the optimal MA positions based on the point-wise channel state information (CSI), without the need for an analytical channel model. To reduce the overhead for CSI acquisition, we also present more efficient learning-based APO algorithms that operate without requiring full point-wise CSI. Finally, we compare the application scenarios of the proposed algorithms and validate their effectiveness with numerical results.

eess.SP↗